Many speech recognition systems use language models to transcribe utterances. Language models are often trained using a set of training data that include examples of language usage. Typically, language models include statistical data about the words or other language units that make up utterances in the training data.